How hysteria over Twitter shadow-banning led to a bizarre congressional hearing

At two occasionally tense congressional hearings Wednesday, members of the Senate Intelligence Committee and the Congressional Committee for Energy and Commerce grilled Twitter CEO Jack Dorsey on a broad range of topics, from concern over the proliferation of bots on Twitter to how the platform deals with hate speech.

But again and again, concerned members of Congress kept returning to one topic: shadow-banning, and the question of whether Twitter is engaging in the practice with regard to conservative voices on the site.

The basic definition of shadow-banning is simple: A member of a given internet community is tacitly blocked or muted to the rest of the community without their knowledge, so that only they can see what they’re posting. Shadow-banning has been around for years — it dates back to early internet forums — but the term has been catapulted into the news this year thanks to a persistent conspiracy theory that Twitter has pointedly and purposefully shadow-banned Republicans who use the site.

The conspiracy theory took root after Twitter made a change to its algorithm that effectively prevented hundreds of thousands of Twitter accounts from being auto-suggested when people used the site’s search function. The change turned out to affect the accounts of many conservative and Republican politicians.

Once this outcome was discovered, in July, Twitter once again changed its algorithm so that the affected accounts displayed normally. But the incident prompted a wave of rumors, right-wing cries about a liberal conspiracy to silence conservatives, and one inflammatory (and false) tweet from President Donald Trump.

And ultimately, outrage over the perceived shadow-banning of conservatives led members of Congress to summon Dorsey to Capitol Hill.

On Wednesday, Republican members of Congress seemed fixated on the idea that Twitter has long displayed unnecessary bias against Republicans through the nebulous practice. During the morning’s Senate hearing, where Dorsey appeared alongside Facebook’s Sheryl Sandberg, and then again during a four-hour afternoon hearing in the House, they repeatedly interrogated Dorsey about whether Twitter’s algorithms have developed inherent biases against conservative users and conservative content.

In response, Dorsey repeatedly explained that the algorithm had no inherent political bias and was simply sorting Twitter content on the basis of numerous behavior signals from the accounts it was reviewing.

Meanwhile, multiple Democrats on the Energy Committee described the proceedings as a waste of time. After outlining the facts around the alleged shadow-banning incident, Rep. Mike Doyle (D-PA) absolved Twitter and Dorsey of any wrongdoing and called the entire exercise of the hearing “a load of crap.” Frank Pallone Jr. (D-NJ) stated that “this hearing appears to be just one more mechanism to raise money and generate outrage. It appears Republicans are desperately trying to rally their base by fabricating a problem that simply does not exist.”

So how did we get here? Is shadow-banning real and has Twitter actually done it? And how did the belief in shadow-banning spur an all-day congressional firing squad?

Let’s break it down — hopefully with more clarity than anyone managed to do on Capitol Hill.

Shadow-banning is an established internet moderator practice in which a user’s posts are muted or hidden without their knowledge, usually because they’re being disruptive in some way. So they get to “participate,” but other people typically can’t see or interact with their posts. The idea is that by shadow-banning someone who’s causing a problem, you avoid riling the person with overt attempts to discipline or restrict them, while also preventing them from bothering others.

The practice used to be quite common, particularly when the internet was mostly comprised of forums and bulletin boards. The term “shadow ban” was reportedly first coined on the seminal forum SomethingAwful way back in 2001, but even before that, forums and bulletin boards across the web gave moderators the power to flag certain users for heightened moderation, or to put them in time-out if they posted too much, too frequently. On mailing lists, moderators would often quietly ban users after fielding multiple complaints about them, or flag certain individuals so their messages would have to be approved by a moderator before they went out to the general populace.

The current understanding of the concept of shadow-banning has been made widely known through Reddit, where it has long been an acceptable method for minimizing bots, spam, trolls, and unfriendly or disruptive users. The modern-day, Reddit-codified version of shadow-banning is to mute a user to the rest of the community without their knowledge.

In the era of social media, however, the old ways born of localized, interpersonal moderation have not only been lost; they’ve also become largely impossible to replicate or maintain on such a large scale. Twitter’s shadow-banning woes sprang from an attempt by the platform to essentially algorithmically replicate that kind of personal moderated space. And like most attempts to algorithmically curate social media these days, this one was destined to encounter a whole lot of problems.

In mid-May, Twitter rolled out several changes to its platform as part of its ongoing quest to “ban the Nazis” and curb site-wide harassment. These changes were specifically designed to combat “troll-like behavior,” meaning that the intended targets were accounts which engage in harassment and “distort” the overall tenor of conversation across the site. Twitter explained this effort, which it pitched as an important step toward “serving healthy conversation,” as follows:

There are many new signals we’re taking in, most of which are not visible externally. Just a few examples include if an account has not confirmed their email address, if the same person signs up for multiple accounts simultaneously, accounts that repeatedly Tweet and mention accounts that don’t follow them, or behavior that might indicate a coordinated attack. We’re also looking at how accounts are connected to those that violate our rules and how they interact with each other.

After the changes, accounts which were algorithmically determined to be exhibiting these and other traits were still viewable on the site — but to see them, you had to search for them and then specifically choose to see hidden results.

In other words, unlike a traditional shadow ban, where you usually can’t access the shadow-banned user’s content at all, Twitter’s version de-listed some accounts and their content from displaying automatically.

This sounds like a pretty useful tool, in theory. But in practice, when Twitter deployed it, the algorithm filtered many accounts belonging to conservative Twitter users. The overall number of affected accounts, which Dorsey announced during Wednesday’s hearing, totaled around 600,000. What caused many observers to balk was that the proportion of conservative accounts that were affected seemed to be very high, and some of the affected accounts belonged to Republican members of Congress.

Here’s how Dorsey explained the mishap in his remarks to the House, which he also tweeted out:

Put simply, Twitter, intending to minimize “troll-like behavior,” wound up inadvertently hiding several Republican-affiliated accounts from the auto-suggest feature in its search function. Many people perceived this event as a concerted attempt by Twitter to silence conservative voices and views on the platform, and reacted with outrage.

In Twitter’s attempt to algorithmically “serve healthy conversation,” it aimed to identify accounts that were potentially harassing others by “repeatedly Tweet[ing] and mention[ing] accounts that don’t follow them”; accounts that were one of many created by the same user; accounts that were participating in coordinated attacks on other users; and accounts that were following users who were doing all of those things. In the process, it appears that Twitter’s algorithm unearthed a rather uncomfortable truth about the way Twitter users connect, and the overlap between trolls and non-trolls on social media.

Users who were affected by the new filter were algorithmically identified as exhibiting the “troll” behavior the algorithm was hunting for, or as engaging — via likes, replies, and follows — with accounts that were exhibiting this behavior. As Dorsey explained on Wednesday, the algorithm also looked at whether an account was followed by a high number of accounts that were exhibiting this behavior. That follower gauge is apparently what led to the accounts of Republican members of Congress not being auto-suggested by Twitter’s search function, which subsequently led Twitter to eventually decide it “wasn’t fair.”

The blunt truth is that in 2018, many of the Twitter accounts that follow and interact with conservative politicians and personalities — including none-the-wiser Republican politicians — are also linked to extremist, racist, and/or white nationalist ideology and have a tendency to engage in abusive and “troll-like” behavior. Thus, Twitter’s algorithm, in its hunt for behavioral patterns, wound up filtering a broad range of users, from fringe extremists and trolls to mainstream Republicans.

Twitter ultimately tweaked its algorithm to restore many of the prominent filtered accounts — while its staff reiterated that those accounts had been affected because of their own behaviors and the behavior of their followers, not because of any inherent political bias.

Despite the explanations, and despite Twitter dialing back the algorithm, the accusations of liberal bias continued to rage. Dorsey wound up fielding questions about the incident from conservative-leaning media and other outlets, and he became the target of outrage from sources ranging from Fox News to Sean Hannity to the well-known conspiracy theorist Alex Jones. Twitter seemed unable to satisfactorily answer congressional concerns, and so Dorsey agreed to meet with the House Energy Committee specifically to address the shadow-ban question and the broader issue of platform bias.

Throughout Wednesday’s hearing, Dorsey referred vaguely to the “behavior signals” that caused the algorithm to affect conservative Twitter. At one point, he described them as “a thousand decision-making criteria that the algorithms use.”

But he didn’t spell out these criteria in terms of the reality that current Republican politics, as embodied by Trump’s own Twitter use, have inevitably become entangled with a lot of horrible, troll-like behavior on social media. (In fact, the hearing itself was attended by a strange coterie of right-wing internet trolls who attempted to troll the session in real life.)

Instead, Dorsey (along with Sandberg during the morning hearing) used a variety of nebulous language throughout the day, which often seemed to frustrate both Republicans and Democrats. Pressed again and again to admit to Twitter showing some type of bias, either systemic or personal, Dorsey consistently demurred. At one point, he sidestepped giving information, when challenged, about whether his own personal political leanings are liberal; at another, he refused to concede that President Trump’s Twitter account might be in violation of Twitter’s general content policies. “We believe strongly in being impartial,” Dorsey said, “and we strive to enforce our rules impartially.”

Members of Congress, in turn, spoke vaguely about instituting government regulations to rein in social media — again, without clarifying what such regulation might look like.

What was made explicitly clear by the end of the very long day was that the entire exercise was something of a wash. Republicans who’d refused to accept Twitter’s initial, consistent explanation of the “shadow ban” appeared no more satisfied at the end of the hearing than they were at the beginning. But Dorsey, the man who has allowed prominent alt-right figures to voice their views on his platform unchecked, was surely not going to stridently defend the downranking of conservative viewpoints, so anyone hoping for that was equally disappointed.

All in all, Dorsey’s day on Capitol Hill offered greater insight into political partisanship than it offered into the inner workings of social media companies. In a cultural moment where the mysterious ways of algorithms are becoming more broadly distrusted by society at large, Twitter’s filtering algorithm seems to have become the latest political football in a game with no winners, and little to show for it other than bruising all around.